Sev1: Snowpipe stopped loading — files land in storage but tables stay empty

#Symptoms

• Dashboards fed by Snowpipe stop updating; files keep arriving in the stage location.
• No errors in the pipeline that writes files, because the failure is on the load side.

#1. Check the pipe status

SELECT SYSTEM$PIPE_STATUS('raw.events.pipe_events');

Key fields in the JSON:

If paused:

ALTER PIPE raw.events.pipe_events SET PIPE_EXECUTION_PAUSED = FALSE;

#2. Check load history for errors

SELECT file_name, status, first_error_message, row_count, last_load_time
FROM TABLE(INFORMATION_SCHEMA.COPY_HISTORY(
  TABLE_NAME => 'RAW.EVENTS.EVENTS',
  START_TIME => DATEADD(hour, -24, CURRENT_TIMESTAMP())))
ORDER BY last_load_time DESC;

#3. Notifications not arriving

S3: the bucket event notification must send ObjectCreated events to the pipe's SQS queue (notification_channel in SHOW PIPES):

aws s3api get-bucket-notification-configuration --bucket my-landing-bucket

Someone adding another notification with Terraform can overwrite the existing configuration, which silently removes Snowpipe's.

Azure: Event Grid subscription to the storage queue and the notification integration must be enabled:

SHOW NOTIFICATION INTEGRATIONS;
DESC NOTIFICATION INTEGRATION ni_events;

az eventgrid event-subscription list --source-resource-id <storage-account-id> -o table

Check the Snowflake service principal still has the Storage Queue Data Contributor role.

Path filters: if producers changed the folder path, the pipe's COPY (stage path) or the notification prefix filter may no longer match.

#4. Storage integration permissions

DESC STORAGE INTEGRATION si_landing;
LIST @raw.events.stage_events;

If LIST fails, the cloud role/consent behind the integration changed (AWS IAM role trust policy, Azure consent or role assignment).

#5. Backfill missed files

After fixing, Snowpipe does not automatically pick up files whose notifications were lost. Refresh loads files staged in the last 7 days that haven't been loaded:

ALTER PIPE raw.events.pipe_events REFRESH;
-- or a narrower path
ALTER PIPE raw.events.pipe_events REFRESH PREFIX = '2026/10/31/';

For older files, use COPY INTO directly. Snowflake's load metadata prevents loading the same file twice for 64 days for COPY; for Snowpipe, check the documentation for its load history window.

Warning

Don't use FORCE = TRUE on COPY INTO during backfill unless you have deduplication downstream. It reloads files already loaded and creates duplicates.

#Prevent it next time

• Alert when lastReceivedMessageTimestamp or table row counts stop moving for N minutes.
• Manage bucket notifications in one place in IaC so they aren't overwritten.
• Validate file format changes with producers through data contracts.
• Monitor COPY_HISTORY for failed files daily.